A flexible load regulation method and system for a virtual power plant with computing power center participation

Abstract

The application discloses a kind of flexibility load regulation and control method and system of computing power center participation virtual power plant, the method includes: constructing computing power center energy consumption portrait model and computing power task feature atlas;Real-time running state data of computing power center is obtained;Based on real-time running state data and computing power task feature atlas, computing power load adjustable potential dynamic estimation is carried out, and the current adjustable load vector of computing power center is obtained;Based on computing power center energy consumption portrait model, computing power task feature atlas, computing power center current adjustable load vector, generate adjustable task candidate pool;Based on computing power center current adjustable load vector and computing power task multidimensional attribute vector, generate flexible load regulation and control strategy candidate set to carry out optimal regulation and control strategy selection, generate target flexible load regulation and control scheme and regulation and control explanation information are issued to computing power center execution terminal.The application significantly improves the regulation and control accuracy, timeliness and long-term adaptability of virtual power plant to computing power center load resources.

Description

Technical Field

[0001] This invention relates to the field of power dispatching and management technology, and more specifically, to a method and system for flexible load control in which a computing center participates in a virtual power plant. Background Technology

[0002] A virtual power plant is a coordinated management system that uses advanced information and communication technologies and software systems to aggregate, manage, and optimize distributed energy resources such as distributed power sources, energy storage systems, and controllable loads. By participating in electricity market transactions and providing grid ancillary services, it achieves efficient utilization of distributed energy resources and safe and stable grid operation. As a crucial load component in the new power system, the computing center's internal server clusters, storage devices, and cooling systems consume enormous amounts of power and have significant flexibility. By incorporating the computing center into the flexible load resource pool of the virtual power plant, the spatiotemporal transfer and power regulation of power load can be achieved without affecting the quality of computing services, providing ancillary services such as peak shaving and frequency regulation for the grid.

[0003] Currently, computing centers participate in grid load regulation mainly through fixed-period power limiting strategies and simple demand response invitation methods. Fixed-period power limiting strategies forcibly reduce the operating power of computing centers by setting pre-set peak-period power limits; demand response invitation methods guide computing centers to adjust their operating power during specific periods by issuing price or incentive signals from the grid side. However, existing computing center load regulation technologies still have the following technical shortcomings:

[0004] First, existing load regulation methods only take the overall power of the computing center as the regulation object, ignoring the deep coupling relationship between the computing task load and the power load inside the computing center. This results in coarse load regulation granularity, making it difficult to distinguish between deferred computing tasks, transferable computing tasks and real-time rigid computing tasks. The assessment of controllable potential is one-sided, and the load regulation accuracy is low.

[0005] Second, existing control systems mostly adopt static power reduction strategies, treating the upper limit of computing center power as a fixed scalar and setting it once. They do not consider the randomness of computing task arrival, the dynamic fluctuation of computing resource utilization, and the real-time constraints of computing service quality. This leads to a mismatch between the control strategy and the actual operating state of the computing center, resulting in strong time lag in the control results and difficulty in adapting to rapid changes in computing load.

[0006] Third, existing methods have a single dimension in modeling computing task attributes. They usually only mark the amount of computing resources required and the priority level of the task, and lack refined modeling of multi-dimensional attributes such as task power consumption characteristics, data dependency strength, cross-center transmission cost, deadline constraints and privacy compliance attributes. This leads to serious homogenization in the selection of control tasks and makes it impossible to provide control schemes with the best cost-effectiveness and the lowest service loss according to the power grid control needs and computing service quality requirements.

[0007] The above shortcomings urgently need to be improved. Summary of the Invention

[0008] To overcome the problems of coarse control granularity, static and rigid control strategies, and insufficient task attribute modeling dimensions in existing computing center load control methods, this invention provides a flexible load control method and system for computing centers participating in virtual power plants.

[0009] The technical solution of this invention is as follows:

[0010] In a first aspect, the present invention provides a method for flexible load regulation of a computing center participating in a virtual power plant, comprising:

[0011] Based on the basic attribute information and historical operation data of the computing center, a model for the energy consumption of the computing center is constructed.

[0012] Based on the basic attribute information and historical execution statistics of computing power tasks, a feature map of computing power tasks is constructed.

[0013] Obtain real-time operational status data from the computing center;

[0014] Based on the real-time operating status data and the computing power task feature map, the adjustable potential of computing power load is dynamically estimated to obtain the current adjustable load vector of the computing power center.

[0015] Based on the computing center energy consumption profile model and the computing task feature map, a multi-dimensional attribute vector of computing tasks is constructed, and an adjustable task candidate pool is generated based on the current adjustable load vector of the computing center.

[0016] Based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of the computing task, the grid regulation matching degree and the computing service quality balance value are calculated, and a candidate set of flexible load regulation strategies is generated.

[0017] Based on the candidate set of flexible load control strategies, the optimal control strategy is selected, the target flexible load control scheme and control explanation information are generated, and then sent to the execution terminal of the computing center.

[0018] As a preferred embodiment of the present invention, the computing center energy consumption profile model includes a basic energy consumption baseline vector of the computing center, a priori distribution of computing load of the computing center, and characteristics of the computing center's operating behavior mode.

[0019] The computing power task feature map includes a set of computing power task nodes, a set of computing resource nodes, a set of data dependency edges between computing power tasks, and a set of adaptation edges between computing power tasks and computing resources.

[0020] As a preferred embodiment of the present invention, the step of dynamically estimating the adjustable potential of computing load based on the real-time operating status data and the computing power task feature map, and obtaining the current adjustable load vector of the computing power center, includes:

[0021] The real-time operating status data is input into a preset computing power load adjustable potential assessment model;

[0022] The adjustable computing load potential assessment model extracts features from the real-time operating status data through a multi-channel feature encoder to obtain a multi-modal operating status feature vector; wherein, the multi-channel feature encoder includes a computing load rate encoding channel, a cooling system power consumption encoding channel, a renewable energy consumption ratio encoding channel, and a grid frequency status encoding channel.

[0023] The multimodal operating state feature vector is input into the task portability evaluation network, and combined with the computing power task association relationship in the computing power task feature map, the current adjustable load vector of the computing power center is obtained. The task portability evaluation network includes a task queue memory unit and a task association propagation unit. The task queue memory unit is used to record the portability of each computing power task, and the task association propagation unit is used to realize the propagation and update of the task portability between associated computing power tasks based on the set of data dependency edges between computing power tasks.

[0024] As a preferred embodiment of the present invention, the computing power load adjustable potential assessment model further includes a task timeliness decay calculation unit. The task timeliness decay calculation unit is used to calculate the task timeliness decay factor based on the time interval between the current scheduling time and the computing power task submission time, and the correlation centrality value of the computing power task in the computing power task feature map during the update of the task portability assessment network, and to perform decay correction on the corresponding task portability in the task queue memory unit according to the task timeliness decay factor.

[0025] As a preferred embodiment of the present invention, the step of constructing a multi-dimensional attribute vector of computing tasks based on the computing center energy consumption profile model and the computing task feature map, and generating an adjustable task candidate pool based on the current adjustable load vector of the computing center, includes:

[0026] Based on the computing power task feature map and the preset computing power task attribute annotation rules, a multi-dimensional attribute vector of computing power tasks is constructed.

[0027] Based on the current adjustable load vector of the computing power center, identify the current adjustable load range and the current adjustable load range of the computing power center;

[0028] Search the computing power task feature map for task nodes that are connected to the current adjustable load range of the computing power center or the current adjustable load range of the computing power center by edge, and generate an adjustable task candidate pool.

[0029] As a preferred embodiment of the present invention, the step of calculating the grid regulation matching degree and the computing power service quality balance value based on the current controllable load vector of the computing power center and the multi-dimensional attribute vector of the computing power task, and generating a candidate set of flexible load regulation strategies, includes:

[0030] Based on the current adjustable load vector of the computing center and the task power consumption level in the multi-dimensional attribute vector of the computing task, the power grid regulation matching degree between the current adjustable load value of the computing center and the task power consumption level is calculated. The power grid regulation matching degree is used to characterize the expected load adjustment amount after the computing task migration is performed.

[0031] Based on the task service quality sensitivity value in the multi-dimensional attribute vector of the computing power task and the computing power center service quality margin value in the current controllable load vector of the computing power center, the computing power service quality balance value is calculated. The computing power service quality balance value is used to characterize the degree of balance between the computing power service quality loss and the power grid control benefit when the computing power center performs load control.

[0032] Based on the power grid regulation matching degree and the computing power service quality balance value, the regulation utility value of each candidate regulation strategy is calculated through a preset regulation utility function;

[0033] Based on the regulation utility value, the regulation strategies in the controllable task candidate pool are sorted in descending order, and the top-ranked regulation strategies by a predetermined number constitute a flexible load regulation strategy candidate set.

[0034] As a preferred embodiment of the present invention, the step of selecting the optimal control strategy based on the candidate set of flexible load control strategies, generating a target flexible load control scheme and control interpretation information, and sending it to the computing center execution terminal includes:

[0035] Based on the candidate set of flexible load control strategies, the optimal control strategy is selected through a preset diversity constraint strategy. The diversity constraint strategy is used to ensure that the control strategies in the candidate set of flexible load control strategies have distributional differences in the dimensions of task type coverage, load adjustment amplitude, and control time scale, thereby generating a target flexible load control scheme.

[0036] Based on the current adjustable load vector of the computing center, the target flexible load control scheme, and the power grid control matching degree, control interpretation information is generated;

[0037] The target flexible load control scheme and the control interpretation information are sent to the computing center execution terminal.

[0038] As a preferred embodiment of the present invention, it further includes:

[0039] Based on the response data of the computing power center to the target flexible load control scheme, the parameters of the computing power load controllability potential assessment model and control strategy are dynamically updated through a preset closed-loop feedback update mechanism.

[0040] As a preferred embodiment of the present invention, the step of dynamically updating the assessment model and control strategy parameters of the adjustable potential of the computing load based on the response data of the computing power center to the target flexible load control scheme through a preset closed-loop feedback update mechanism includes:

[0041] Collect response data from the computing center to the target flexible load control scheme;

[0042] Calculate the control effect evaluation index based on the response data;

[0043] Based on the aforementioned regulation effect evaluation index, the parameters of the computing power load controllability potential evaluation model and the task portability evaluation network are dynamically updated.

[0044] Based on the aforementioned regulation effect evaluation index, the weight coefficients in the regulation utility function and the constraint thresholds in the diversity constraint strategy are dynamically updated.

[0045] Secondly, the present invention provides a flexible load control system for computing power centers participating in virtual power plants, and a method for implementing flexible load control of computing power centers participating in virtual power plants using any of the above-mentioned schemes, comprising:

[0046] The energy consumption profiling unit is used to construct an energy consumption profiling model of the computing center based on the basic attribute information and historical operation data of the computing center.

[0047] The feature map construction unit is used to construct a feature map of computing power tasks based on the basic attribute information and historical execution statistics of computing power tasks.

[0048] The operation status acquisition terminal is used to collect real-time operation status data of the computing center;

[0049] The load adjustable potential assessment unit is used to dynamically estimate the adjustable potential of computing load based on the real-time operating status data, and obtain the current adjustable load vector of the computing center.

[0050] The task attribute modeling unit is used to construct a multi-dimensional attribute vector of computing power tasks based on the computing power task feature map and the preset computing power task attribute annotation rules.

[0051] The task candidate pool generation unit is used to construct a multi-dimensional attribute vector of computing tasks based on the computing center energy consumption profile model and the computing task feature map, and to generate an adjustable task candidate pool based on the current adjustable load vector of the computing center.

[0052] The strategy generation unit is used to calculate the grid regulation matching degree and the balance value of computing service quality based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of the computing task, and generate a set of flexible load regulation strategy candidates to select the optimal regulation strategy, and generate the target flexible load regulation scheme and regulation explanation information.

[0053] The computing center execution terminal is used to receive and execute the target flexible load control scheme and display the control explanation information.

[0054] According to the above-described solution, the beneficial effects of this invention are as follows:

[0055] 1. By constructing a computing center energy consumption profile model and a computing task feature map, personalized energy consumption benchmarks and prior constraints are provided for the dynamic estimation of the adjustable potential of computing load and the precise control of flexible load. A structured task association basis is provided for the correlation propagation of the portability of computing tasks and the precise retrieval of the adjustable task candidate pool.

[0056] 2. By acquiring real-time operational status data of the computing center, multi-dimensional operational information such as computing load rate, cooling system power consumption, renewable energy consumption ratio, and grid frequency status are incorporated into the analysis scope. This provides rich input features that reflect the actual operational status of the computing center for the computing load controllable potential assessment model, enabling the controllable potential estimation to distinguish between rigid computing load and flexible computing load.

[0057] 3. By constructing a multi-dimensional attribute vector of computing power tasks and generating a controllable task candidate pool based on the current controllable load vector of the computing power center, a fine characterization of multi-dimensional attributes such as task power consumption level, service quality sensitivity value, data dependency strength and cross-center transmission cost is realized, as well as highly correlated retrieval with power grid control requirements.

[0058] 4. By calculating the grid regulation matching degree and the equilibrium value of computing power service quality, and generating a candidate set of flexible load regulation strategies, the precise matching of grid regulation demand and computing power center operation constraints, as well as the optimal balance between computing power service quality loss and grid regulation benefit are achieved.

[0059] 5. By selecting the optimal control strategy from the candidate set of flexible load control strategies, the target flexible load control scheme and control explanation information are generated and sent to the execution terminal of the computing center. While ensuring that the control strategies have distributional differences in terms of task type coverage, load adjustment range and control time scale, the computing center operator is provided with transparent information such as description of control reasons, controllable task prompts and expected control benefits, which significantly improves the computing center operator's willingness to participate and trust in the virtual power plant control system. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating a method for flexible load control in a virtual power plant involving a computing center, according to an embodiment of the present invention.

[0061] Figure 2 for Figure 1 A detailed flowchart of step S4 is shown below;

[0062] Figure 3 for Figure 1 A detailed flowchart of step S5 is shown below;

[0063] Figure 4 for Figure 1 A detailed flowchart of step S6 is shown below;

[0064] Figure 5 for Figure 1 A detailed flowchart of step S7 is shown below;

[0065] Figure 6 This is a flowchart illustrating a flexible load control method for a computing center participating in a virtual power plant, as described in another embodiment of the present invention.

[0066] Figure 7 for Figure 6 A detailed flowchart of step S8 is shown below;

[0067] Figure 8 This is a schematic diagram of the structure of a flexible load control system in which a computing center participates in a virtual power plant, according to one embodiment of the present invention.

[0068] Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0069] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0070] It should be noted that the terms "comprising" and "having," and any variations thereof, in the specification and claims of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0071] like Figure 1 As shown, one embodiment of the present invention provides a flexible load control method for a computing center participating in a virtual power plant, comprising:

[0072] S1. Based on the basic attribute information and historical operation data of the computing center, construct an energy consumption profile model for the computing center.

[0073] Specifically, the process involves: acquiring basic attribute information and historical operational data of the computing center. Basic attribute information includes the computing center's geographical location, server room size, cooling method, and network bandwidth configuration. Historical operational data includes historical power curve records, historical computing load rate records, historical cooling system power consumption records, historical renewable energy consumption ratio records, and historical task execution time period distribution information. Based on this information, an energy consumption profile model of the computing center is constructed. This model includes a basic energy consumption baseline vector, a priori distribution of computing load, and operational behavior pattern characteristics. The basic energy consumption baseline vector is obtained by weighted averaging of the average power in each time period from the historical power curve records. The priori distribution of computing load is obtained through time-period Bayesian inference of historical computing load rate records. The operational behavior pattern characteristics are obtained through cluster analysis of historical task execution time period distribution information and historical cooling system power consumption records.

[0074] This step constructs an energy consumption profile model of computing centers by collecting basic attribute information and historical operation data. It establishes a personalized representation framework that reflects the individual energy consumption characteristics and operation patterns of computing centers, overcoming the shortcomings of existing technologies that ignore individual differences of computing centers and adopt uniform power reduction strategies. This provides personalized energy consumption benchmarks and prior constraints for subsequent dynamic estimation of the adjustable potential of computing load and precise control of flexible load.

[0075] S2. Construct a feature map of computing power tasks based on the basic attribute information and historical execution statistics of computing power tasks.

[0076] Specifically, this involves: acquiring basic attribute information and historical execution statistics of computing power tasks, and constructing a feature map of computing power tasks based on these information. The basic attribute information of computing power tasks includes the text description of the task, the dataset information involved, the data dependencies involved, and the historical execution statistics. The feature map includes a set of computing power task nodes, a set of computing resource nodes, a set of data dependency edges between computing power tasks, and a set of adaptation edges between computing power tasks and computing resources. Each computing power task node in the set contains an identifier, name, business type, and priority level attribute. Each computing resource node in the set contains an identifier, type, and priority level attribute. The source rated power and computing resource geographical location attributes; each edge in the data dependency relationship edge set contains the source computing power task node identity identifier, the target computing power task node identity identifier, and the dependency strength weight. The dependency strength weight is calculated based on the task execution order constraints and the influence coefficient of the output data volume of the preceding task on the execution time of the subsequent task in the historical execution statistics; each edge in the adaptation relationship edge set contains the computing power task node identity identifier, the computing resource node identity identifier, and the adaptation degree weight. The adaptation degree weight is calculated based on the matching degree between the computing power task resource requirement characteristics and the computing resource performance characteristics, as well as the average energy consumption efficiency of the computing resource in executing the computing power task in the historical execution statistics.

[0077] This step constructs a feature map of computing tasks by using basic attribute information and historical execution statistics of computing tasks. It establishes a structured network of relationships between computing tasks and between computing tasks and computing resources. This overcomes the shortcomings of isolated scheduling of computing tasks and lack of task association paths in existing technologies. It provides a structured task association foundation for the correlation propagation of the portability of computing tasks and the accurate retrieval of the controllable task candidate pool.

[0078] S3. Obtain real-time operating status data of the computing center.

[0079] Specifically, the real-time operational status data of the computing center is acquired through an operational status acquisition terminal, and this data is preprocessed. The real-time operational status data includes computing load rate, cooling system power consumption, renewable energy consumption ratio, grid frequency status, and computing task queue status. The operational status acquisition terminal is deployed at the execution terminal of the computing center and includes a computing load monitoring module, a cooling power consumption recording module, an energy status acquisition module, and a grid status monitoring module. The computing load monitoring module records the current CPU utilization and memory utilization of each computing resource in the computing center, generating the computing load rate. The cooling power consumption recording module records the current real-time operating power of the computing center's cooling system, generating the cooling system power consumption. The energy status acquisition module records the proportion of renewable energy power generation currently consumed by the computing center to the total power consumption, generating the renewable energy consumption ratio. The grid status monitoring module records the current grid frequency measurement value at the computing center's grid connection point, generating the grid frequency status. Preprocessing includes normalization of computing load rate, removal of outliers and logarithmic transformation of cooling system power consumption, interval truncation of renewable energy consumption ratio, standardization of grid frequency status, and timestamp alignment of real-time operating status data to ensure that each modal data has a corresponding relationship under a unified time base.

[0080] This step acquires real-time operational status data of the computing center by deploying a multimodal operational status acquisition terminal at the execution terminal of the computing center. This breaks through the limitation of traditional load control that only collects total power data, and obtains multi-dimensional operational data that reflects the internal energy consumption composition of the computing center. This provides multi-dimensional input features for the computing load controllability potential assessment model, enabling the controllability potential estimation to distinguish between rigid computing loads and elastic computing loads.

[0081] S4. Based on real-time operational status data and computing power task characteristic maps, dynamically estimate the adjustable potential of computing power load to obtain the current adjustable load vector of the computing power center. For example... Figure 2 As shown, the specific steps include:

[0082] S401. Input the real-time operating status data into the preset computing load adjustable potential assessment model.

[0083] S402, the computing power load controllable potential assessment model extracts features from real-time operating status data using a multi-channel feature encoder to obtain multi-modal operating status feature vectors. The multi-channel feature encoder includes a computing power load rate encoding channel, a cooling system power consumption encoding channel, a renewable energy consumption ratio encoding channel, and a grid frequency status encoding channel.

[0084] The computing load rate encoding channel normalizes and embeds the computing load rate into an embedding vector to obtain a computing load rate feature vector. Specifically, the computing load rate is normalized to the range of zero to one, the normalized value is evenly divided into a preset number of load interval buckets, and the bucket index is mapped to an embedding space of a preset dimension through an embedding layer to obtain the computing load rate feature vector.

[0085] The cooling system power consumption encoding channel performs logarithmic transformation and bucketing encoding on the cooling system power consumption to obtain a cooling power consumption feature vector. Specifically, the cooling system power consumption is transformed by natural logarithm to eliminate the influence of extreme values, the logarithmically transformed value is evenly divided into a preset number of power consumption interval buckets, and the bucket index is mapped to the cooling power consumption feature vector through an embedding layer.

[0086] The renewable energy consumption ratio coding channel performs interval coding and attention weight calculation on the renewable energy consumption ratio to obtain an energy state feature vector. Specifically, the renewable energy consumption ratio is divided into a preset number of consumption level intervals, each level interval is encoded as an interval identifier vector, the importance weight of each consumption level interval is calculated through a self-attention mechanism, and the interval identifier vectors are weighted and summed based on the importance weights to obtain the energy state feature vector.

[0087] The power grid frequency state coding channel performs standardized coding and fully connected transformation on the power grid frequency state to obtain the power grid state feature vector. Specifically, the deviation value between the power grid frequency state and the rated frequency is standardized, and the standardized deviation value is mapped to a preset dimension through a fully connected layer and a nonlinear activation function to obtain the power grid state feature vector.

[0088] The multimodal operating state feature vector is obtained by concatenating the feature vectors of computing power load rate, cooling power consumption, energy status, and power grid status.

[0089] S403. Input the multimodal operating state feature vector into the task portability assessment network, and combine it with the computing power task correlation in the computing power task feature map to obtain the current adjustable load vector of the computing power center. The task portability assessment network includes a task queue memory unit, a task correlation propagation unit, and a task timeliness decay calculation unit.

[0090] The task queue memory unit is used to record the portability of each computing task.

[0091] The task association propagation unit is used to propagate and update the portability of a task among associated computing tasks based on the set of data dependency edges between computing tasks. Specifically, when the task queue memory unit updates the portability of a certain computing task, the task association propagation unit queries the set of subsequent associated tasks for that computing task based on the set of data dependency edges in the computing task feature graph. Based on the dependency strength weights in the data dependency edge set, it propagates the change in the portability of the current computing task proportionally to the subsequent associated tasks, updating the estimated portability of the subsequent associated tasks. The propagation update formula is:

[0092] ;

[0093] in, This is an estimate of the portability of computing task j at time t; This is the task association propagation coefficient, with a value ranging from 0 to 1; The dependency strength weight of the data dependency edge pointing from computing task i to computing task j; This represents the change in the portability of computing task i after the current scheduling interaction.

[0094] The task timeliness decay calculation unit is used during the task portability assessment network update process to calculate the task timeliness decay factor based on the time interval between the current scheduling time and the task submission time, and the correlation centrality value of the task in the task feature map. The formula for calculating the task timeliness decay factor is:

[0095] ;

[0096] in, Let be the time-decrease factor of computing task i at time t; The moment when computing task i is submitted to the task queue; For time intervals; The baseline parameter for the time-dependent decay rate is determined by the characteristics of the computing center's operational behavior mode in the computing center's energy consumption profile model. The correlation centrality value of computing task i in the computing task feature map is calculated by the weighted average of the degree centrality, betweenness centrality and eigenvector centrality of the computing task. It represents the core degree of the computing task in the task system. The higher the core degree of the computing task, the closer the time decay factor is to 1, that is, the less suitable it is for migration.

[0097] The task timeliness decay calculation unit is also used to decay and correct the portability of the corresponding tasks in the task queue memory unit according to the task timeliness decay factor:

[0098] ;

[0099] in, This is an estimate of the portability of computing task i after time decay correction.

[0100] The output vector of the task queue memory unit after task association propagation and time-related decay correction is used as the current adjustable load vector of the computing center.

[0101] This step overcomes the limitations of traditional load regulation, which relies solely on total power data, by fusing multi-modal operating status data through a multi-channel feature encoder. The task association propagation unit utilizes the computing power task feature map to propagate and update the portability of tasks among associated tasks, ensuring that the controllable potential estimation conforms to the inherent logic of the task structure. Furthermore, the task timeliness decay calculation unit introduces a timeliness decay factor based on time interval and task association centrality, enabling the controllable potential estimation to reflect the time-sensitive characteristics of computing power tasks, significantly improving the accuracy and timeliness of dynamic estimation of the controllable potential of computing power load.

[0102] S5. Based on the energy consumption profile model of the computing center and the feature map of computing tasks, construct a multi-dimensional attribute vector of computing tasks, and generate an adjustable task candidate pool based on the current adjustable load vector of the computing center. For example... Figure 3 As shown, the specific steps include:

[0103] S501. Based on the computing power task feature map and the preset computing power task attribute annotation rules, construct a multi-dimensional attribute vector for computing power tasks. This multi-dimensional attribute vector includes the task power consumption level, the task service quality sensitivity value, the task data dependency strength value, the task deadline margin value, and the task cross-center transmission cost value.

[0104] The task power consumption level is calculated based on the historical average power consumption from the historical execution statistics of the computing power task. Specifically, the task power consumption level is equal to the index value of the historical average power consumption mapped to a preset discrete level range; the higher the historical average power consumption, the higher the task power consumption level.

[0105] The task service quality sensitivity value is calculated based on a comprehensive assessment of the task's deadline constraint level, task service type priority, and task execution interruption recovery cost. The calculation formula is:

[0106] ;

[0107] in, Sensitivity value for service quality of the task; The deadline constraint level for computing power tasks is determined by annotations from subject matter experts or business operators. The highest deadline constraint level; Priority level for the business type to which the computing power task belongs; This is the highest business priority level. This is an estimated cost to recover from an interruption in the execution of a computing task. This is the estimated maximum recovery cost. , , Let be the weight coefficients of each item, and let the sum of the weight coefficients equal 1.

[0108] The task data dependency strength value is calculated based on the ratio between the output data of the preceding task and the input data of the current task. The higher the ratio, the higher the task data dependency strength value, and the greater the transmission cost required for cross-center migration.

[0109] The task deadline margin is calculated based on the time interval between the task deadline timestamp and the current time. The larger the time interval, the higher the task deadline margin and the larger the schedulable space.

[0110] The cross-center transmission cost of a task is calculated based on the task data dependency strength value and the network bandwidth cost coefficient between computing centers.

[0111] S502. Based on the current adjustable load vector of the computing center, identify the current adjustable load range and the current adjustable load range of the computing center. Specifically, compare the estimated portability of each dimension in the current adjustable load vector of the computing center with a preset portability threshold. Computing tasks with estimated portability values ​​higher than the portability threshold and corresponding to tasks that can be delayed in execution constitute the task sources for the current adjustable load range of the computing center. Computing tasks with estimated portability values ​​higher than the portability threshold and corresponding to tasks that can be executed in advance constitute the task sources for the current adjustable load range of the computing center.

[0112] S503. Retrieve computing task nodes in the computing power task feature map that are connected to the current adjustable load range or the current adjustable load range of the computing power center through matching edges, and generate an adjustable task candidate pool. During the retrieval, computing task nodes that conflict with the current operating status of the computing power center are excluded to ensure that the candidate tasks have a reasonable matching relationship with the current resource capacity of the computing power center.

[0113] This step provides a multidimensional decision-making basis for the control strategy by refining the multidimensional attribute vector of computing power tasks; by generating a candidate pool of controllable tasks based on the current controllable load vector of the computing center and the feature map of computing power tasks, the high correlation between candidate tasks and control requirements is ensured.

[0114] S6. Based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of computing tasks, calculate the grid control matching degree and the balance value of computing service quality, and generate a candidate set of flexible load control strategies. For example... Figure 4 As shown, the specific steps include:

[0115] S601. Based on the current adjustable load vector of the computing center and the task power consumption level in the multi-dimensional attribute vector of the computing task, calculate the grid regulation matching degree between the current adjustable load value of the computing center and the task power consumption level. The current adjustable load value of the computing center is calculated as the weighted average of the migration degree estimates corresponding to the current adjustable load range or the current adjustable load range of the computing center in the current adjustable load vector of the computing center. The weights are determined by the correlation centrality value of the computing task in the computing task feature map. The grid regulation matching degree is used to characterize the expected load adjustment amount after executing the computing task migration, and the calculation formula is:

[0116] ;

[0117] in, The power grid regulation matching degree between computing center u and computing task e, with a value range between 0 and 1; This represents the current adjustable load value of the computing center u; The power consumption level for computing task e; This represents the natural exponential function. When the grid regulation matching degree is close to 0.5, it indicates that the task power consumption level is best matched with the controllable load value of the computing center, and the expected load adjustment amount is closest to the regulation demand.

[0118] S602. Based on the task service quality sensitivity value in the multi-dimensional attribute vector of computing power tasks and the computing power center service quality margin value in the current controllable load vector of the computing power center, calculate the computing power service quality equilibrium value. The computing power center service quality margin value is obtained by calculating the norm distance between the basic energy consumption baseline vector of the computing power center and the current controllable load vector of the computing power center in the computing power center energy consumption profile model. It characterizes the balance between the computing power service quality loss and the grid regulation benefits when the computing power center performs load regulation. The formula for calculating the computing power service quality equilibrium value is:

[0119] ;

[0120] in, This is the quality of service balance value between computing power center u and computing power task e, and its value ranges from 0 to 1; The sensitivity value of the quality of service for computing task e; The service quality margin value for computing center u; This indicates that the maximum value is taken during the operation. When the computing power service quality balance value is close to 1, it means that the sensitivity of the task service quality and the service quality margin of the computing power center are most balanced. When the computing power center performs regulation, it will not suffer serious service quality loss due to the task being too sensitive, nor will it lose regulation value due to excessive conservatism.

[0121] S603. Based on the balance between power grid regulation matching degree and computing power service quality, calculate the regulation utility value of each candidate regulation strategy through a preset regulation utility function. The formula for calculating the regulation utility value is as follows:

[0122] ;

[0123] in, Let $e$ be the effect of computing power task $e$ on computing power center $u$. , , Let be the weight coefficients of each item, satisfying that the sum of the weight coefficients equals 1; The carbon emission reduction potential value of computing task e is calculated by weighting the proportion of renewable energy consumption during the execution period of the computing task with the task's power consumption level. The first term of the regulation utility function... The first term is the power grid control matching utility term, which reaches its maximum value when the power grid control matching degree is 0.5, encouraging the selection of tasks that best match the controllable load value of the computing center; the second term... For the utility term of balancing computing power service quality, it encourages the selection of tasks with the least service quality loss; the third term For potential carbon emission reduction utility items, encourage the selection of tasks that can maximize the absorption of renewable energy.

[0124] S604. Based on the control utility value, sort the control strategies in the controllable task candidate pool in descending order, and select the control strategies with the highest control utility value to form a flexible load control strategy candidate set.

[0125] This step overcomes the rigidity of existing simple power reduction strategies by jointly optimizing the grid control matching degree and the computing power service quality balance value. The generated control strategy achieves comprehensive optimization in three dimensions: grid matching degree, service quality balance value and carbon emission reduction potential. This ensures that the virtual power plant obtains the expected load regulation amount while avoiding the risk of service quality default caused by excessive control by the computing power center.

[0126] S7. Based on the candidate set of flexible load control strategies, the optimal control strategy is selected, the target flexible load control scheme and control explanation information are generated, and then sent to the execution terminal of the computing center. For example... Figure 5 As shown, the specific steps include:

[0127] S701. Based on the candidate set of flexible load control strategies, the optimal control strategy is selected through a preset diversity constraint strategy. The diversity constraint strategy ensures that the control strategies in the candidate set of flexible load control strategies have distributional differences in three dimensions: task type coverage, load adjustment amplitude, and control time scale, thereby generating a target flexible load control scheme.

[0128] The specific implementation of the diversity constraint strategy is as follows: In the candidate set of flexible load control strategies, the control strategy with the highest control utility value is first selected as the first target flexible load control scheme; then, among the remaining control strategies, control strategies that have an overlap with the selected control strategy in the set of computing power task node identity identifiers exceeding a preset overlap threshold are excluded, control strategies whose load adjustment amplitude difference with the selected control strategy is less than a preset amplitude difference threshold are excluded, and control strategies with the same control time scale as the selected control strategy are excluded; among the remaining control strategies after exclusion, the control strategy with the highest control utility value is selected as the next target flexible load control scheme; the above process is repeated until the number of selected target flexible load control schemes reaches the preset recommended number.

[0129] S702. Based on the current controllable load vector of the computing center, the multi-dimensional attribute vector of the computing task of the target flexible load control scheme, and the grid control matching degree, generate control explanation information. This control explanation information includes a text describing the control reasons, a text prompting the current controllable tasks of the computing center, and a text describing the expected control benefits.

[0130] The generation rules for the regulation reason description text are as follows: If the grid regulation matching degree is between 0.4 and 0.6, the regulation reason description text "The load regulation amount of this regulation scheme is highly matched with the grid demand, and it is suitable for precise power regulation" is generated; if the grid regulation matching degree is less than 0.4, the regulation reason description text "This regulation scheme prioritizes ensuring the quality of computing power services and is suitable for light load optimization" is generated; if the grid regulation matching degree is greater than 0.6, the regulation reason description text "This regulation scheme has great load regulation potential and is suitable for deep peak shaving response" is generated.

[0131] The rule for generating the prompt text for the currently controllable tasks of the computing power center is as follows: extract the intersection of the set of identity identifiers of computing power task nodes in the target flexible load control scheme and the set of computing power tasks in the current controllable load vector of the computing power center whose estimated mobility is higher than the threshold. Sort the names of computing power tasks in the intersection in descending order of correlation centrality value to generate the prompt text "This scheme prioritizes scheduling the following controllable computing power tasks: List of computing power task names".

[0132] The generation rule for the expected regulation benefit description text is as follows: Based on the multi-dimensional attribute vector of the computing power task of the target flexible load regulation scheme and the current controllable load vector of the computing power center, predict the expected load adjustment amount and the expected carbon emission reduction amount after the implementation of the regulation scheme, and generate the benefit description text "After the implementation of this scheme, the power of the computing power center will be adjusted by the expected adjustment amount of megawatts, and the carbon emissions will be reduced by the expected reduction amount of kilograms".

[0133] S703. The target flexible load control scheme and control interpretation information are sent to the computing center execution terminal. The computing center execution terminal presents the task migration instructions, computing resource reconfiguration instructions, and power adjustment instructions in the target flexible load control scheme in the form of task scheduling instructions, and presents the control interpretation information in the form of interpretation pop-ups or monitoring panels.

[0134] This step ensures the multidimensional diversity of the control strategy set through a diversity constraint strategy, avoiding resource conflicts caused by homogeneous control; by generating control interpretation information, it enhances the transparency of the virtual power plant control system and the trust of the computing center operator, enabling the operator to understand the power grid operation logic behind the control and increasing their willingness to participate.

[0135] This embodiment's method for flexible load regulation of a computing center participating in a virtual power plant, by constructing a computing center energy consumption profile model and a computing task feature map, provides personalized energy consumption benchmarks and prior constraints for subsequent dynamic estimation of the controllable potential of computing load and precise flexible load regulation. It also provides a structured task association foundation for the correlation propagation of the portability of computing tasks and the precise retrieval of the controllable task candidate pool. By acquiring real-time operating status data of the computing center, it incorporates multi-dimensional operating information such as computing load rate, cooling system power consumption, renewable energy consumption ratio, and grid frequency status into the analysis scope, overcoming the limitations of traditional methods. Existing technologies rely solely on total power data and neglect energy consumption characteristics at the task level. This paper addresses the shortcomings of existing technologies, which rely solely on total power data and neglect energy consumption characteristics at the task level. It provides rich input features reflecting the actual operating status of the computing center, enabling the estimation of adjustable potential to distinguish between rigid and flexible computing loads. By using a pre-defined computing load adjustable potential assessment model based on real-time operating status data and computing task feature maps, the paper obtains the current adjustable load vector of the computing center, achieving a dynamic and refined estimation of the portability of each computing task within the center. This overcomes the limitations of existing technologies. The static and rigid estimation of controllable potential in computing centers fails to reflect the correlation between tasks and the evolution of computing load, thus hindering the dynamic adaptation of control measures to rapid changes in computing load. By constructing a multi-dimensional attribute vector of computing tasks based on a computing task feature map and pre-defined computing task attribute annotation rules, and generating a candidate pool of controllable tasks based on the current controllable load vector of the computing center, this approach achieves refined characterization of multi-dimensional attributes such as task power consumption level, quality of service sensitivity value, data dependency strength value, and cross-center transmission cost value, as well as highly correlated retrieval with power grid control requirements. This overcomes the limitations of existing technologies in terms of task attribute dimension... This addresses the shortcomings of the previous technology, such as the single-dimensionality and coarse generation of candidate task pools. By calculating the grid regulation matching degree and the balance value of computing service quality based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of computing tasks, and generating a candidate set of flexible load regulation strategies, it achieves accurate matching between grid regulation demand and computing center operation constraints, as well as the optimal balance between computing service quality loss and grid regulation benefits. This overcomes the shortcomings of the existing technology, such as the mismatch between the regulation scheme and the actual operating state of the computing center caused by the simple power reduction strategy, and the strong time lag of the regulation results. The generated regulation strategy has both the expected regulation effect and avoids service quality default.By selecting the optimal control strategy from the candidate set of flexible load control strategies, a target flexible load control scheme and control explanation information are generated and distributed to the execution terminal of the computing center. This ensures that the control strategies have distributional differences in task type coverage, load adjustment amplitude, and control time scale, while providing the computing center operator with transparent information such as descriptions of the control reasons, controllable task prompts, and expected control benefits. This overcomes the shortcomings of existing technologies, such as severe homogenization of control schemes and lack of interpretability in the control process, significantly improving the computing center operator's willingness to participate and their trust in the virtual power plant control system.

[0136] like Figure 6 As shown, in one embodiment, the flexible load regulation method for a computing center participating in a virtual power plant according to the present invention further includes:

[0137] S8. Based on the response data of the computing power center to the target flexible load control scheme, the parameters of the computing power load controllability potential assessment model and control strategy are dynamically updated through a preset closed-loop feedback update mechanism. For example... Figure 7 As shown, the specific steps include:

[0138] S801. Collect response data from the computing center to the target flexible load control scheme. This response data includes the actual load adjustment amount, actual task migration time, actual computing power service quality indicators, and subjective feedback from the computing center operator. The subjective feedback from the computing center operator includes their subjective rating of the difficulty of the control scheme and their subjective rating of satisfaction with the control measures.

[0139] S802. Calculate the control effect evaluation indicators based on response data. Among them, the control effect evaluation indicators include the deviation between the actual load adjustment amount and the expected load adjustment amount, the deviation between the actual task migration time and the expected task migration time, and the subjective satisfaction score of the computing center operator.

[0140] The expected load adjustment is directly given by the grid control matching degree, that is, the expected load adjustment is equal to the grid control matching degree multiplied by the task power consumption level of the computing power task.

[0141] The formula for calculating the deviation between the actual load adjustment and the expected load adjustment is as follows:

[0142] ;

[0143] in, This represents the actual load adjustment.

[0144] The expected task migration time is obtained by jointly estimating the characteristics of the computing center's operating behavior pattern in the computing center energy consumption profile model and the cost of task cross-center transmission.

[0145] The formula for calculating the deviation between the actual task migration time and the expected task migration time is as follows:

[0146] ;

[0147] in, The actual time required for the migration of the task. The expected time required for task migration.

[0148] The subjective satisfaction score of the computing center operator is directly obtained from the operator's subjective satisfaction with the regulation, and the value ranges from 1 to 5.

[0149] S803. Based on the regulation effect evaluation index, the parameters of the multi-channel feature encoder and the task transferability evaluation network in the computing power load controllability potential evaluation model are dynamically updated using a preset model parameter update algorithm. The model parameter update algorithm employs a stochastic gradient descent algorithm based on the deviation value, with the loss function being:

[0150] ;

[0151] in, The loss function value for the model assessing the adjustable potential of computing load; The weighting coefficient for the deviation in task migration time; This is the weighting coefficient for the satisfaction deviation. The subjective satisfaction rating of the computing center operator is calculated. By minimizing the loss function value, the parameters of the multi-channel feature encoder and the task transferability assessment network are updated using the backpropagation algorithm, making the output of the computing load controllability potential assessment model more closely reflect the actual response of the computing center.

[0152] S804. Based on the regulation effect evaluation index, the weight coefficients in the regulation utility function and the constraint thresholds in the diversity constraint strategy are dynamically updated using a preset strategy parameter update algorithm. The strategy parameter update rule is as follows: when the subjective satisfaction score of the computing center operator consistently exceeds the preset high satisfaction threshold, the weight coefficient of the power grid regulation matching utility term in the regulation utility function is increased. Reduce the weighting coefficient of the carbon emission reduction potential utility term. Meanwhile, the overlap threshold and amplitude difference threshold in the diversity constraint strategy are relaxed to provide a control scheme that is more in line with the current comfort zone of the computing center; when the subjective satisfaction score of the computing center operator is consistently lower than the preset low satisfaction threshold, the weight coefficient of the power grid control matching utility term is reduced. Increase the weighting coefficient of the equilibrium utility term for computing power service quality. Weighting coefficients of the carbon emission reduction potential utility term At the same time, the constraint thresholds in the diversity constraint strategy are tightened to provide a more service quality assurance and more diverse control scheme.

[0153] This step, through a closed-loop feedback update mechanism, enables the adjustable potential assessment model of computing load to continuously optimize the accuracy of adjustable potential estimation based on the actual response of the computing center. It also enables the control strategy parameters to adaptively adjust the control style based on the satisfaction feedback of the computing center operator, thereby realizing the self-evolution capability of the control system and significantly improving the long-term adaptability of the system and the willingness of the computing center to continue participating.

[0154] like Figure 8 As shown, the present invention also provides a flexible load control system for a computing center participating in a virtual power plant, comprising:

[0155] The energy consumption profiling unit is used to construct an energy consumption profiling model of the computing center based on the basic attribute information and historical operation data of the computing center.

[0156] The feature map construction unit is used to construct a feature map of computing power tasks based on the basic attribute information and historical execution statistics of computing power tasks.

[0157] The operation status acquisition terminal is deployed in the execution terminal of the computing center and is used to collect real-time operation status data of the computing center.

[0158] The load controllability potential assessment unit is used to dynamically estimate the computing load controllability potential based on real-time operating status data and obtain the current controllable load vector of the computing center.

[0159] The task attribute modeling unit is used to construct a multi-dimensional attribute vector of computing tasks based on the computing task feature map and the preset computing task attribute annotation rules.

[0160] The task candidate pool generation unit is used to construct a multi-dimensional attribute vector of computing tasks based on the computing center energy consumption profile model and computing task feature map, and to generate an adjustable task candidate pool based on the current adjustable load vector of the computing center.

[0161] The strategy generation unit is used to calculate the grid regulation matching degree and the balance value of computing service quality based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of computing tasks, and to generate a set of candidate flexible load regulation strategies for optimal regulation strategy selection, and generate target flexible load regulation scheme and regulation interpretation information.

[0162] The closed-loop feedback update unit is used to dynamically update the parameters of the computing load controllability potential assessment model and control strategy based on the response data of the computing center to the target flexible load control scheme.

[0163] The computing center execution terminal is used to receive and execute target flexible load control schemes and display control explanation information, as well as collect real-time operating status data and response data.

[0164] The flexible load control system for the computing center participating in the virtual power plant provided in this application embodiment can execute the above method embodiment. Its implementation principle and technical effect are similar, and will not be repeated here.

[0165] Figure 9 This is a schematic diagram of the electronic device used in the present invention to realize the flexible load control method of the computing center participating in the virtual power plant.

[0166] Electronic devices may include processors, memory, communication buses and communication interfaces, and may also include computer programs stored in memory and capable of running on the processor, such as flexible load control programs for virtual power plants in which computing centers participate.

[0167] In some embodiments, the processor can be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control unit of the electronic device, connecting various components of the device through various interfaces and lines. It executes programs or modules stored in memory (e.g., executing flexible load control programs for virtual power plants in computing centers) and calls data stored in memory to perform various functions and process data within the electronic device.

[0168] The memory includes at least one type of readable storage medium, including flash memory, portable hard drives, multimedia cards, card-type memory (e.g., SD or DX memory), magnetic storage, magnetic disks, optical disks, etc. In some embodiments, the memory can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory can be an external storage device of the electronic device, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory can include both internal and external storage units of the electronic device. The memory can be used not only to store application software and various types of data installed on the electronic device, such as the code of a flexible load control program for a computing center participating in a virtual power plant, but also to temporarily store data that has been output or will be output.

[0169] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. The bus is configured to enable communication between memory and at least one processor.

[0170] The communication interface is used for communication between the aforementioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0171] Figure 9 Only electronic devices with relevant components are shown; it will be understood by those skilled in the art that... Figure 9 The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0172] For example, a power supply, although not shown, may also include a power source (such as a battery) to power various components. Preferably, the power supply can be logically connected to at least one processor via a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power sources, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be elaborated further here.

[0173] It should be understood that the embodiments are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0174] The computing center stored in the memory of electronic devices participates in the flexible load control program of the virtual power plant, which is a combination of multiple computer programs. For the specific implementation method of the processor for these computer programs, please refer to [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0175] Furthermore, if the modules / units integrated into an electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, a computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0176] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following functions: Figure 1 The flowchart illustrates the process of a computing center participating in the flexible load control method of a virtual power plant.

[0177] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0178] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0179] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0180] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

[0181] The present invention has been described above with reference to the accompanying drawings. Obviously, the implementation of the present invention is not limited to the above-described manner. Any improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A method for flexible load regulation involving a computing center in a virtual power plant, characterized in that, include: Based on the basic attribute information and historical operation data of the computing center, a computing center energy consumption profile model is constructed. Based on the basic attribute information and historical execution statistics of computing power tasks, a feature map of computing power tasks is constructed. Obtain real-time operational status data from the computing center; Based on the real-time operating status data and the computing power task feature map, the adjustable potential of computing power load is dynamically estimated to obtain the current adjustable load vector of the computing power center. Based on the computing center energy consumption profile model and the computing task feature map, a multi-dimensional attribute vector of computing tasks is constructed, and an adjustable task candidate pool is generated based on the current adjustable load vector of the computing center. Based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of the computing task, the grid regulation matching degree and the computing service quality balance value are calculated, and a candidate set of flexible load regulation strategies is generated. Based on the candidate set of flexible load control strategies, the optimal control strategy is selected, the target flexible load control scheme and control explanation information are generated, and then sent to the execution terminal of the computing center.

2. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 1, characterized in that, The computing center energy consumption profile model includes the basic energy consumption baseline vector of the computing center, the prior distribution of computing load of the computing center, and the characteristics of the computing center's operating behavior mode. The computing power task feature map includes a set of computing power task nodes, a set of computing resource nodes, a set of data dependency edges between computing power tasks, and a set of adaptation edges between computing power tasks and computing resources.

3. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 1, characterized in that, The process of dynamically estimating the adjustable potential of computing load based on the real-time operating status data and the computing power task feature map, and obtaining the current adjustable load vector of the computing power center, includes: The real-time operating status data is input into a preset computing power load adjustable potential assessment model; The adjustable computing load potential assessment model extracts features from the real-time operating status data through a multi-channel feature encoder to obtain a multi-modal operating status feature vector; wherein, the multi-channel feature encoder includes a computing load rate encoding channel, a cooling system power consumption encoding channel, a renewable energy consumption ratio encoding channel, and a grid frequency status encoding channel. The multimodal operating state feature vector is input into the task portability evaluation network, and combined with the computing power task association relationship in the computing power task feature map, the current adjustable load vector of the computing power center is obtained. The task portability evaluation network includes a task queue memory unit and a task association propagation unit. The task queue memory unit is used to record the portability of each computing power task, and the task association propagation unit is used to realize the propagation and update of the task portability between associated computing power tasks based on the set of data dependency edges between computing power tasks.

4. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 3, characterized in that, The computing load adjustable potential assessment model also includes a task timeliness decay calculation unit. The task timeliness decay calculation unit is used to calculate the task timeliness decay factor based on the time interval between the current scheduling time and the computing task submission time, and the correlation centrality value of the computing task in the computing task feature map during the update of the task portability assessment network, and to decay and correct the corresponding task portability in the task queue memory unit according to the task timeliness decay factor.

5. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 1, characterized in that, The process of constructing a multi-dimensional attribute vector for computing tasks based on the computing center energy consumption profile model and the computing task feature map, and generating an adjustable task candidate pool based on the current adjustable load vector of the computing center, includes: Based on the computing power task feature map and the preset computing power task attribute annotation rules, a multi-dimensional attribute vector of computing power tasks is constructed. Based on the current adjustable load vector of the computing power center, identify the current adjustable load range and the current adjustable load range of the computing power center; Search the computing power task feature map for task nodes that are connected to the current adjustable load range of the computing power center or the current adjustable load range of the computing power center by edge, and generate an adjustable task candidate pool.

6. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 3, characterized in that, The process of calculating the grid regulation matching degree and the equilibrium value of computing power service quality based on the current controllable load vector of the computing power center and the multi-dimensional attribute vector of the computing power task, and generating a candidate set of flexible load regulation strategies, includes: Based on the current adjustable load vector of the computing center and the task power consumption level in the multi-dimensional attribute vector of the computing task, the power grid regulation matching degree between the current adjustable load value of the computing center and the task power consumption level is calculated. The power grid regulation matching degree is used to characterize the expected load adjustment amount after the computing task migration is performed. Based on the task service quality sensitivity value in the multi-dimensional attribute vector of the computing power task and the computing power center service quality margin value in the current controllable load vector of the computing power center, the computing power service quality balance value is calculated. The computing power service quality balance value is used to characterize the degree of balance between the computing power service quality loss and the power grid control benefit when the computing power center performs load control. Based on the power grid regulation matching degree and the computing power service quality balance value, the regulation utility value of each candidate regulation strategy is calculated through a preset regulation utility function; Based on the regulation utility value, the regulation strategies in the controllable task candidate pool are sorted in descending order, and the top-ranked regulation strategies by a predetermined number constitute a flexible load regulation strategy candidate set.

7. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 6, characterized in that, The process of selecting the optimal control strategy based on the candidate set of flexible load control strategies, generating a target flexible load control scheme and control explanation information, and sending it to the computing center execution terminal includes: Based on the candidate set of flexible load control strategies, the optimal control strategy is selected through a preset diversity constraint strategy. The diversity constraint strategy is used to ensure that the control strategies in the candidate set of flexible load control strategies have distribution differences in the dimensions of task type coverage, load adjustment amplitude, and control time scale, thereby generating a target flexible load control scheme. Based on the current adjustable load vector of the computing center, the target flexible load control scheme, and the power grid control matching degree, control interpretation information is generated; The target flexible load control scheme and the control interpretation information are sent to the computing center execution terminal.

8. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 7, characterized in that, Also includes: Based on the response data of the computing power center to the target flexible load control scheme, the parameters of the computing power load controllability potential assessment model and control strategy are dynamically updated through a preset closed-loop feedback update mechanism.

9. The method for flexible load regulation of a computing center participating in a virtual power plant according to claim 8, characterized in that, The response data from the computing power center to the target flexible load control scheme is used to dynamically update the computing power load controllability potential assessment model and control strategy parameters through a preset closed-loop feedback update mechanism, including: Collect response data from the computing center to the target flexible load control scheme; Calculate the control effect evaluation index based on the response data; Based on the aforementioned regulation effect evaluation index, the parameters of the computing power load controllability potential evaluation model and the task portability evaluation network are dynamically updated. Based on the aforementioned regulation effect evaluation index, the weight coefficients in the regulation utility function and the constraint thresholds in the diversity constraint strategy are dynamically updated.

10. A flexible load control system for a virtual power plant with computing power center participation, used to execute the flexible load control method for a virtual power plant with computing power center participation as described in any one of claims 1 to 9, characterized in that, include: The energy consumption profiling unit is used to construct an energy consumption profiling model of the computing center based on the basic attribute information and historical operation data of the computing center. The feature map construction unit is used to construct a feature map of computing power tasks based on the basic attribute information and historical execution statistics of computing power tasks. The operation status acquisition terminal is used to collect real-time operation status data of the computing center; The load adjustable potential assessment unit is used to dynamically estimate the adjustable potential of computing load based on the real-time operating status data, and obtain the current adjustable load vector of the computing center. The task attribute modeling unit is used to construct a multi-dimensional attribute vector of computing power tasks based on the computing power task feature map and the preset computing power task attribute annotation rules. The task candidate pool generation unit is used to construct a multi-dimensional attribute vector of computing tasks based on the computing center energy consumption profile model and the computing task feature map, and to generate an adjustable task candidate pool based on the current adjustable load vector of the computing center. The strategy generation unit is used to calculate the grid regulation matching degree and the balance value of computing service quality based on the current controllable load vector of the computing center and the multi-dimensional attribute vector of the computing task, and generate a set of flexible load regulation strategy candidates to select the optimal regulation strategy, and generate the target flexible load regulation scheme and regulation explanation information. The computing center execution terminal is used to receive and execute the target flexible load control scheme and display the control explanation information.

Images